學術出版;;Publications

Permanent URI for this communityhdl:11536/1

收集所有交通大學研究人員與學生個人或團體所發表之各項學術著作，涵蓋期刊論文、教師專書、學位論文等項。;;Collections of Publications of NCTU Researchers and Students Like Journal Papers and Thesis.

Browse

🏛️國立陽明交通大學機構典藏

Search Results

Now showing 1 - 10 of 66

DSNet: An Efficient CNN for Road Scene Segmentation
(2019-01-01)
Chen, Ping-Rong
;
Hang, Hsueh-Ming
;
Chan, Sheng-Wei
;
Lin, Jing-Jhih
;
交大名義發表
Road scene understanding is a critical component in an autonomous driving system. Although the deep learning-based road scene segmentation can achieve very high accuracy, its complexity is also very high for developing real-time applications. It is challenging to design a neural net with high accuracy and low computational complexity. To address this issue, we investigate the advantages and disadvantages of several popular CNN architectures in terms of speed, storage and segmentation accuracy. We start from the Fully Convolutional Network (FCN) with VGG, and then we study ResNet and DenseNet. Through detailed experiments, we pick up the favorable components from the existing architectures and at the end, we construct a light-weight network architecture based on the DenseNet. Our proposed network, called DSNet, demonstrates a real-time testing (infer encing) ability (on the popular GNI platform) and it maintains an accuracy comparable with most previous systems. We test our system on several datasets including the challenging Cityscapes dataset (resolution of 1024 x512) with an mIoU of about 69.1 % and runtime of 0.0147 second per image on a single GTX 1080Ti. We also design a more accurate model but at the price of a slower speed, which has an mIoU of about 72.6 % on the CamVid dataset.
LEARNED IMAGE COMPRESSION WITH SOFT BIT-BASED RATE-DISTORTION OPTIMIZATION
(2019-01-01)
Alexandre, David
;
Chang, Chih-Peng
;
Peng, Wen-Hsiao
;
Hang, Hsueh-Ming
;
交大名義發表
This paper introduces the notion of soft bits to address the rate-distortion optimization for learning-based image compression. Recent methods for such compression train an autoencoder end-to-end with an objective to strike a balance between distortion and rate. They are faced with the zero gradient issue due to quantization and the difficulty of estimating the rate accurately. Inspired by soft quantization, we represent quantization indices of feature maps with differentiable soft bits. This allows us to couple tightly the rate estimation with context-adaptive binary arithmetic coding. It also provides a differentiable distortion objective function. Experimental results show that our approach achieves the state-of-the-art compression performance among the learning-based schemes in terms of MS-SSIM and PSNR.
INCORPORATING LUMINANCE, DEPTH AND COLOR INFORMATION BY A FUSION-BASED NETWORK FOR SEMANTIC SEGMENTATION
(2019-01-01)
Hung, Shang-Wei
;
Lo, Shao-Yuan
;
Hang, Hsueh-Ming
;
交大名義發表
;
電子工程學系及電子研究所
Semantic segmentation has made encouraging progress due to the success of deep convolutional networks in recent years. Meanwhile, depth sensors become prevalent nowadays; thus, depth maps can be acquired more easily. However, there are few studies that focus on the RGB-D semantic segmentation task. Exploiting the depth information effectiveness to improve performance is a challenge. In this paper, we propose a novel solution named LDFNet, which incorporates Luminance, Depth and Color information by a fusion-based network. It includes a sub-network to process depth maps and employs luminance images to assist the depth information in processes. LDFNet outperforms the other state-of-art systems on the Cityscapes dataset, and its inference speed is faster than most of the existing networks. The experimental results show the effectiveness of the proposed multi-modal fusion network and its potential for practical applications.
NCTU-GTAV360: A 360 degrees Action Recognition Video Dataset
(2019-01-01)
Ardianto, Sandy
;
Hang, Hsueh-Ming
;
電機學院
;
College of Electrical and Computer Engineering
Despite many action recognition video datasets available right now, none of them are in the spherical projection. NCTU-GTAV360 is a new 360 degrees action recognition video dataset captured from a game, Grand Theft Auto V (GTA V). The spherical video is obtained by stitching 24 views from various angles and combining them into a video. The benefit of using 360 degrees cameras is that it can capture the entire surroundings using one single camera. We captured 200 locations within the Los Santos city (city name in the GTA V). This dataset should benefit researchers working on the spherical images, particularly the human action recognition research using machine learning or deep learning technique, which requires a large amount of training data and the associated ground-truth.
Multi-Class Lane Semantic Segmentation using Efficient Convolutional Networks
(2019-01-01)
Lo, Shao-Yuan
;
Hang, Hsueh-Ming
;
Chan, Sheng-Wei
;
Lin, Jing-Jhih
;
交大名義發表
Lane detection plays an important role in a self-driving vehicle. Several studies leverage a semantic segmentation network to extract robust lane features, but few of them can distinguish different types of lanes. In this paper, we focus on the problem of multi-class lane semantic segmentation. Based on the observation that the lane is a small-size and narrow-width object in a road scene image, we propose two techniques, Feature Size Selection (FSS) and Degressive Dilation Block (DD Block). The FSS allows a network to extract thin lane features using appropriate feature sizes. To acquire fine-grained spatial information, the DD Block is made of a series of dilated convolutions with degressive dilation rates. Experimental results show that the proposed techniques provide obvious improvement in accuracy, while they achieve the same or faster inference speed compared to the baseline system, and can run at real-time on high-resolution images.
Multi-View and Multi-Modal Action Recognition with Learned Fusion
(2018-01-01)
Ardianto, Sandy
;
Hang, Hsueh-Ming
;
電機學院
;
College of Electrical and Computer Engineering
In this paper, we study multi-modal and multi-view action recognition system based on the deep-learning techniques. We extended the Temporal Segment Network with additional data fusion stage to combine information from different sources. In this research, we use multiple types of information from different modality such as RGB, depth, infrared data to detect predefined human actions. We tested various combinations of these data sources to examine their impact on the final detection accuracy. We designed 3 information fusion methods to generate the final decision. The most interested one is the Learned Fusion Net designed by us. It turns out the Learned Fusion structure has the best results but requires more training.
DEPTH MAP ENHANCEMENT ON RGB-D VIDEO CAPTURED BY KINECT V2
(2018-01-01)
Lin, Ke-Yu
;
Hang, Hsueh-Ming
;
交大名義發表
;
National Chiao Tung University
The quality of depth map is one key factor contributing to the quality of 3D video and virtual reality (VR) rendering. In this study, we use RGB-D camera (Microsoft Kinect for Windows v2) to capture the color sequences and depth sequences as our system inputs. The captured depth map contains various noises and artifacts in addition to the occlusion regions. We use the color sequences in both spatial domain and time domain to improve the quality of the depth map. Our main contributions are alignment between color and depth images and reducing artifacts in the reflection regions. Several techniques are adopted, modified, and re-designed such as moving object compensation, unreliable depth pixel detection, and locally adaptive depth pixel refinement algorithm. The experimental results show that the quality of the depth map is significantly improved.
Learning-based saliency model with depth information
(2015-01-01)
Ma, Chih-Yao
;
Hang, Hsueh-Ming
;
電子工程學系及電子研究所
;
Department of Electronics Engineering and Institute of Electronics
Most previous studies on visual saliency focused on two-dimensional (2D) scenes. Due to the rapidly growing three-dimensional (3D) video applications, it is very desirable to know how depth information affects human visual attention. In this study, we first conducted eye-fixation experiments on 3D images. Our fixation data set comprises 475 3D images and 16 subjects. We used a Tobii TX300 eye tracker (Tobii, Stockholm, Sweden) to track the eye movement of each subject. In addition, this database contains 475 computed depth maps. Due to the scarcity of public-domain 3D fixation data, this data set should be useful to the 3D visual attention research community. Then, a learning-based visual attention model was designed to predict human attention. In addition to the popular 2D features, we included the depth map and its derived features. The results indicate that the extra depth information can enhance the saliency estimation accuracy specifically for close-up objects hidden in a complex-texture background. In addition, we examined the effectiveness of various low-, mid-, and high-level features on saliency prediction. Compared with both 2D and 3D state-of-the-art saliency estimation models, our methods show better performance on the 3D test images. The eye-tracking database and the MATLAB source codes for the proposed saliency model and evaluation methods are available on our website.
Online Multiclass Passive-Aggressive Learning on a Fixed Budget
(2017-01-01)
Wu, Chung-Hao
;
Hsi, Wei-Chen
;
Lu, Henry Horng-Shing
;
Hang, Hsueh-Ming
;
統計學研究所
This paper presents a budgetary learning algorithm for online multiclass classification. Based on the multiclass passive-aggressive learning with kernels, we introduce a dual perspective that gives rise to the proposed budgetary algorithm. Basically, the proposed algorithm limits the amount of data in use and fully exploits the available data on hand through optimization. The algorithm has both constant time and space complexities and thus can avoid the curse of kernelization. Experimental results with open datasets show that the proposed budgetary algorithm is competitive with state-of-the-art algorithms.
FAST TEMPORAL PREDICTION SELECTION FOR H.264/AVC SCALABLE VIDEO CODING
(2009-01-01)
Lin, Hung-Chih
;
Hang, Hsueh-Ming
;
Peng, Wen-Hsiao
;
資訊工程學系
;
電子工程學系及電子研究所
In this paper, we propose a fast algorithm that selects the temporal prediction type for the dyadic hierarchical prediction structure in scalable video coding (SVC). We make use of the strong correlations in the large block partitions to eliminate the unnecessary computations for bi-directional prediction. Moreover, based upon the information of an 8x8 partition, either forward or backward prediction is skipped for its smaller block partitions. Comparing to the JSVM 9.11, our method saves the encoding time from 50% to 60% for a number of test videos over a typical range of coding bit-rates and its coding penalty is negligible.*

學術出版;;Publications

Browse

Filters

Settings

Sort By

Results per page

🏛️國立陽明交通大學機構典藏

Search Results